Re: [fluid-dev] Propose changes for drum channels support

[Chris Moeller]

Further to this, I propose a change to the bank number handling. While I 
already saw some changes under way to use MSB or LSB or both depending 
on the MIDI standard, I think it's better to combine the approach.

Since SoundFont hardware devices expect the bank change commands to only 
be in the MSB, and those correspond to the bank numbers encoded in the 
SoundFont banks themselves, I think FluidSynth should behave 
accordingly. Namely, it should read SoundFont bank numbers scaled to 
MSB. The player code should handle both MSB and LSB, and in case of XG 
MIDI data setting the bank LSB, the preset search fallback should search 
for a bank number with the LSB masked off in the event that a MSB+LSB 
match is not found. This is what I currently do in my modified 
FluidSynth which I link into foo_midi.

Of course, I'm not sure what to do about the bank LSB field. While the 
SoundFont 2.x file format does support a 16-bit bank field, I don't 
think any editors actually allow setting the bank to higher than 128, 
the drum bank.

Although the above seems to be fine for when I'm using a largely General 
MIDI preset with AnotherXG as the first bank loaded, as a fallback. In 
fact, it even takes care of XG style drum mapping, since bank 16256, or 
MSB 127, is filled with drum instruments in that bank. In fact, my 
higher level fallback bank, SGP v2.01, even has drum instruments in bank 
127.

So, I have changed the defined drum bank to 16256, and the default 
SoundFont loader also multiplies file bank numbers by 128 so they're in 
the MSB, or sets it to 16256 if it's 128. This nicely handles XG drum 
channel settings, and falls bank on the normal bank 128 drum kits in 
case no SoundFonts with bank 127 drum kits are loaded. If a SoundFont 
has both bank 127 and bank 128 drum kits, the bank 127 kits will be 
found first. Maybe this needs to be optimized so that if bank 127 
defines drum kits, corresponding instruments in bank 128 will be 
ignored. Hmm...